On Thursday 10 January 2002 17:22, Matthew Walker wrote: > Where I work, we're at the point of needing to maintain two database > servers, one inhouse, and the other at another physical location. We > need to replicate between the servers, to keep them both up to date, as > both will be getting updated. This leaves us with the problem of > resolving primary key conflicts if the servers should ever lose touch > with eachother for a while. > > We're looking at InnoDB, and wondering if it would be able to resolve > those sorts of issues using transactions. > > Input please?
We looked closely at "native" MySQL replication, but after a few questions about the handling of auto increment fields (we use them like row IDs all over the place), we decided that MySQL's replication was broken for our application. We have more than one live database and they replicate to each other. We write our own replication system using our own checkpointing and code. We start with the update logs (we're not happy about the future removal of the update logs because of this). I've looked through the MySQL source to see if I can figure out enough of the format of the binary log to see if I can make some Perl code that will parse it. It is really easy to use the update log since all SQL end with a semicolon as the last character on a line. Since Perl tends to be line-oriented, it is easy to find this. Here's what we do: - we run a special server process on each database. This replicator server will spool update log files on demand. - on each machine, we run a special client process. This process connects to the replicator server (not MySQL's), gets the update log information and puts it into the local database. If necessary, we can rewrite any SQL (we don't). The server process puts a special checkpoint comment between each SQL statement. The checkpoint comments has a special header and trailer so that we can recognize it. It contains the name of the file that is currently being spooled and the byte offset in that file. This information is stored on the client side in a special checkpoint file. If the client process needs to be restarted or the network drops or something causes replication to quit, it will restart where it left off based on the checkpoint file. This functionality is basically identical to the native MySQL replication. The reasons for using the update log and not native MySQL replication are these: - we can have more that two servers coupled together. - we use auto increment fields all over. With native mySQL replication, these fields are filled in with values on the source side. Thus, we get collisions on the target side. We have no need for the row IDs to match on each machine as every row has another unique ID field. We do need the row IDs for purging old data and other things that do not rely on the unique ID field. - we can do SQL rewriting if we really need to. These problems led us to write our own replication. If you have the luxury of redesigning your database or of designing with MySQL's native replication in mind (we did not), then I would use the native version. Best, Kyle --------------------------------------------------------------------- Before posting, please check: http://www.mysql.com/manual.php (the manual) http://lists.mysql.com/ (the list archive) To request this thread, e-mail <[EMAIL PROTECTED]> To unsubscribe, e-mail <[EMAIL PROTECTED]> Trouble unsubscribing? Try: http://lists.mysql.com/php/unsubscribe.php
